Achieving Long-Term Surveillance

in VigilNetTian He, Pascal Vicaire, Ting Yan, Qing Cao, Gang Zhou, Lin Gu, Liqian Luo, Radu Stoleru,

John A. Stankovic, Tarek F. Abdelzaher

Department of Computer ScienceUniversity of VirginiaCharlottesville, USA

Motivating Application: Battlefield Surveillance

Other Applications

Wildlife Monitoring

Alarm System

Flock Protection

Border Surveillance

Our Solution: VigilNet

MICA2 / MICAz / XSM Motes

MAC Sensor Drivers

Routing Power Mngt 1

SignalFiltering

TimeSync

Localization GroupMngt

PowerMng 2

Power Mngt 3

ProgrammingAbstractions

TargetClassification

Velocity / TrajectoryInference

Physical

Data-Link

Routing

Middleware

Application

0%

1%

1%99%

Initialization Sleep Event Process

Communication Surveillance

Focus of This Presentation: Power Consumption No power management => 4 days lifetime!

99% of energy consumed waiting for potential targets!

EnergyDistribution

0%

0%2%

2%

Initialization Sleep Event Process

Communication Surveillance

Focus of This Presentation: Power Consumption Power management => 10 months lifetime!

Lifetime x 75

98% of energy consumed in sleep mode!

EnergyDistribution

98%

State of the Art

Topics: Hardware Energy Scavenging Topology Control Sensing Coverage Predefined

Scheduling Data Aggregation Etc…

Practicality? Performance in Real

Deployments? Applicability to

Surveillance System? Combination of

Schemes?

Power Management in VigilNet Turning nodes off as often and as long as possible. Questions:

When to turn nodes off (to save power)? When to wake nodes up (to optimize system performance)? What are the tradeoffs?

Combination of four schemes: Node level power management. Group level power management. Network level power management. On-demand wakeup.

Group Level: Sentry Selection Redundant Coverage!

Group Level: Sentry Selection Redundant Coverage! => Sentry Selection

Group Level: Sentry Selection Load Balancing?

Group Level: Sentry Selection Load Balancing? => Sentry Rotation

Group Level: Sentry Selection Tradeoff: Detection Latency versus Density

Probability ofTarget Detected

Within First1,000m

Number of Nodes inArea 100m x 1,000m

10010 1,000

Area

1,000m 100m

Radius=20mRadius=8mRadius=2m

50 125 500

Sentry Level: Duty Cycle Scheduling Target Takes Time To Go Through the Network.

Sentry Level: Duty Cycle Scheduling Target Takes Time To Go Through the Network.

=> Duty Cycle Scheduling

Sentry Level: Duty Cycle Scheduling Putting It All Together

Sentry Level: Duty Cycle Scheduling Tradeoff: Detection Latency Versus Duty Cycle

Area

1,000m 100m

Probability ofTarget Detected

Within First1,000m

Duty Cycle

40%

100%

0% 20%

1000 Nodes, V=10m/s1000 Nodes, V=30m/s

Network Level: Tripwire Scheduling Exploiting Knowledge About the Target

Network Level: Tripwire Scheduling Exploiting Knowledge About the Target

Network Level: Tripwire Scheduling Tripwire partition based on distance to a base

On-Demand Wakeup

Wakeup

Wakeup PathTo Base StationWakeup Nodes

For FutureDetection

Detection

Details of Wakeup Operation

Sleeping Node: Wakeup x% of the Time

Wakeup Operation: Send Message with Long Preamble

Toggle Period

Sleep 1% Wakeup Sleep

Preamble length = TogglePeriod * BitRate SYNC Bytes DATA CRC

Evaluation by Third Party: Test Field

Mote Field

300m X 200m, 200 motes

Evaluation by Third Party:Interactive Display

Evaluation by Third Party:Detection, Classification, and Tracking

1.Initial Detection

2.Classification

3.Periodic updates

Average Localization Error: 6.24mAverage Velocity Error: 6%

Lifetime Evaluation: Hybrid Simulation

0

10

20

30

40

50

60

70

Time (seconds)

En

erg

y(m

w)

Sentry

NonSentry

Initialization Duration = 5 minutes

Surveillance Duration = 1day

Without system rotation:NonSentry Life Time: 250 daysSentry LifeTime: 7 days

Key Results: Lifetime

Lifetime No Power Management => 4 Days + Sentry Selection and Rotation => 28 Days + Duty Cycle Scheduling => 5 Months

(12.5% Duty Cycle) + Tripwire Service => 10 Months

(16 Tripwires, ¼ Awake) Tracking Performance Penalty

~ 3 to 5 Seconds 0

50

100

150

200

250

300

350

No PM + Sentry + Duty Cycle + Tripwire

Lifetime (days)

Key Results: Detection Performance Penalty ~ 3 to 5 Seconds

Summary

Successfully integrate 4 power management strategies into real system.

Analytical model and extensive simulation to predict system performance under various configurations.

Practical feasibility of tracking system using XSM2s with 10 months lifetime.

My Webpage: www.cs.virginia.edu/~pv9f

Tian’s Webpage:www.cs.umn.edu/~tianhe

Research Group Webpage:www.cs.virginia.edu/~control

Questions?